Application of Machine Learning in Estimating California Bearing Ratio from Soil Index Properties in Kenya

Résumé

The California Bearing Ratio (CBR) is an important civil and transportation engineering test. It is normally carried out to assess soil's bearing capacity and strength for road pavement and foundation construction. The test, however, is both time-consuming and labour-intensive, resulting in significant delays during the construction process, ultimately leading to financial losses due to the high cost typically associated with construction projects. As a potential solution to this issue, an investigation is conducted into the application of artificial intelligence (AI) and machine learning (ML) techniques for accurately forecasting CBR values. Three models were used in the study, namely, the random forest model, linear regression model, and extreme gradient boosting (XGBoost) model. These models were employed to forecast CBR values based on several soil index properties. These properties included particle size distribution (i.e., percentage of soil passing through the sieve of diameter 0.425mm and 0.075mm), liquid limit (LL), plasticity index (PI), maximum dry density (MDD), plastic limit (PL), and optimum moisture content (OMC). A dataset containing these soil properties and corresponding CBR values for soils was obtained from the University of Nairobi civil engineering laboratory. The models were then trained on 80% of the data and tested on 20%. Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Coefficient of determination (R²) were used to evaluate the accuracy of the predictions. The findings showed that XGBoost was the most accurate model with the lowest MAE, MSE, and RMSE, and the highest R², making it the preferred model for predicting CBR